RESEARCH ARTICLE Open Access

The role of resting-state functional MRI forclinical preoperative language mappingVinodh A. Kumar1* , Islam M. Heiba1, Sujit S. Prabhu2, Melissa M. Chen1, Rivka R. Colen3, Angela L. Young1,Jason M. Johnson1, Ping Hou4, Kyle Noll5, Sherise D. Ferguson2, Ganesh Rao2, Frederick F. Lang2,Donald F. Schomer1 and Ho-Ling Liu4

Abstract

Background: Task-based functional MRI (tb-fMRI) is a well-established technique used to identify eloquent cortex,but has limitations, particularly in cognitively impaired patients who cannot perform language paradigms. Resting-state functional MRI (rs-fMRI) is a potential alternative modality for presurgical mapping of language networks thatdoes not require task performance. The purpose of our study is to determine the utility of rs-fMRI for clinicalpreoperative language mapping when tb-fMRI is limited.

Methods: We retrospectively reviewed 134 brain tumor patients who underwent preoperative fMRI languagemapping. rs-fMRI was post-processed with seed-based correlation (SBC) analysis, when language tb-fMRI waslimited. Two neuroradiologists reviewed both the tb-fMRI and rs-fMRI results. Six neurosurgeons retrospectivelyrated the usefulness of rs-fMRI for language mapping in their patients.

Results: Of the 134 patients, 49 cases had limited tb-fMRI and rs-fMRI was post-processed. Two neuroradiologistsfound rs-fMRI beneficial for functional language mapping in 41(84%) and 43 (88%) cases respectively; Cohen’skappa is 0.83, with a 95% confidence interval (0.61, 1.00). The neurosurgeons found rs-fMRI “definitely” useful in 26cases (60%) and “somewhat” useful in 13 cases (30%) in locating potential eloquent language centers of clinicalinterest. Six unsuccessful rs-fMRI cases were due to: head motion (2 cases), nonspecific functionality connectivityoutside the posterior language network (1 case), and an unknown system instability (3 cases).

Conclusions: This study is a proof of concept that shows SBC rs-fMRI may be a viable alternative for clinicallanguage mapping when tb-fMRI is limited.

Keywords: Functional connectivity, Resting-state fMRI, Seed-based correlation, Regional homogeneity

BackgroundFunctional MRI (fMRI) is a well-known modality oftenused in neurosurgical oncology for preoperative plan-ning to facilitate safe, maximal surgical resection of tu-mors located in eloquent brain areas, including those inlanguage areas [1, 2]. Language mapping with task-basedfMRI (tb-fMRI) requires a combination of multiple

language paradigms to generate reliable and accurate ac-tivation of language networks [1]. tb-fMRI also requireshighly trained personnel to determine patient cognitionand to choose suitable language paradigms. Thus, tb-fMRI can be challenging in brain tumor patients whomay be cognitively impaired and are unable to performrequired tasks.Resting-state fMRI (rs-fMRI) is an alternative modality

for preoperative mapping of language and motor net-works and has the potential to overcome some limita-tions of tb-fMRI [3]. rs-fMRI does not depend on

© The Author(s). 2020 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you giveappropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate ifchanges were made. The images or other third party material in this article are included in the article's Creative Commonslicence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commonslicence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtainpermission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to thedata made available in this article, unless otherwise stated in a credit line to the data.

* Correspondence: vakumar@mdanderson.org1Department of Neuroradiology, The University of Texas MD AndersonCancer Center, Houston, TX, USAFull list of author information is available at the end of the article

Kumar et al. Cancer Imaging (2020) 20:47 https://doi.org/10.1186/s40644-020-00327-w

http://crossmark.crossref.org/dialog/?doi=10.1186/s40644-020-00327-w&domain=pdfhttp://orcid.org/0000-0002-8322-1233http://creativecommons.org/licenses/by/4.0/http://creativecommons.org/publicdomain/zero/1.0/mailto:vakumar@mdanderson.org

patient performance, but rather on the detection ofspontaneous changes in low-frequency blood oxygen-ation level-dependent (BOLD) oscillations to identify as-sociated brain areas while the patient is at rest [4].Further, rs-fMRI can be successfully performed whenthe patient is asleep or even under anesthesia [5]. Tieet al. showed in healthy right-handed individuals, thatlanguage networks obtained from rs-fMRI revealedhighly similar overlap with language networks obtainedfrom tb-fMRI, especially in the left frontal and temporal/parietal regions [6]. These results were also reproducedin patients with brain tumors and epilepsy [7]. Inaddition, Rosazza et al. [8] reported a significant correl-ation of the results of rs-fMRI seed-based correlation(SBC) and independent component analysis (ICA) in 40healthy individuals.To the best of our knowledge, no previous studies

have examined the use of SBC rs-fMRI for clinical lan-guage mapping in a large series of brain tumor patients.The purpose of our study is to determine the utility ofSBC rs-fMRI for preoperative language mapping whentb-fMRI is limited.

MethodsSubjectsOur institutional review board approved this retrospect-ive study (PA19–0417). A total of 134 patients withbrain tumors underwent both tb-fMRI and rs-fMRI forpreoperative language mapping between September 1,2017 and March 31, 2019. Eighty-five patients under-went successful tb-fMRI language mapping, and, thus,rs-fMRI post-processing was not performed. In theremaining 49 patients, rs-fMRI was processed; in the lat-ter, either the patient could not perform the tb-fMRIparadigms, the patient performed the tb-fMRI paradigmspoorly, the results of the tb-fMRI activation specific tothe language area near tumor was equivocal or there wasexcessive patient motion. In these cases, tb-fMRI was de-fined as “limited”. We collected the clinical and fMRIdata for each of these 49 patients.

Imaging techniquesMRI acquisitionStructural MRI and fMRI were performed using a 3 TGE MR750 scanner (GE Healthcare, Waukesha, Wiscon-sin) with an 8-channel head coil. Functional images wereacquired using a T2*-weighted gradient-echo echo-planar imaging sequence (repetition time/echo time =2000/25 ms, matrix size = 64 × 64, field of view = 24 × 24cm, 32 slices, slice thickness = 4 mm with no gap).Thirty-two slices were acquired per dynamic to coverthe entire brain. High-resolution T2-weighted FLAIRand 3D spoiled gradient-echo T1-weighted sequenceswere acquired for anatomic reference. A six-minute rs-

fMRI acquisition was obtained prior to language tb-fMRI in all cases. During this acquisition, the patientwas asked to do the following: close their eyes, not fallasleep, clear their mind, and keep their head still.

Tb-fMRI paradigmsThe following pertains to the 49 cases in which the tb-fMRI results were limited. In 46 English speakingpatients, three fMRI language paradigms were adminis-tered in English: letter fluency, category fluency, andsentence completion. There were three patients who didnot speak English. The category fluency and object nam-ing paradigms were administered in Mandarin in onepatient and in Arabic in another patient. The objectnaming paradigm alone was administered in one Arabicspeaking patient.All paradigms used alternating active and control

blocks. Before tb-fMRI, all patients underwent practicetrials according to our standard preoperative mappingguidelines to ensure that the patient could perform thetasks correctly. During fMRI acquisition, each paradigmwas displayed using an MRI-compatible 32-in. liquidcrystal display, and oral instructions were providedthrough an intercom.

Workflow and processing pipeline (tb-fMRI & rs-fMRI)Task-based (tb) fMRI data were processed by using theDynaSuite Neuro software, version 3.0 (Invivo, Philips,Gainesville, Florida). Image pre-processing for fMRI in-cluded motion correction and spatial smoothing with a4-mm full width at half-maximum Gaussian kernel. Afunctional activation map was created using the correl-ation analysis of the task paradigms convolved with a ca-nonical hemodynamic response function and the signalintensity time course for each voxel. Statistical thresh-olds ranging from corrected p < 10− 6 to 10− 2 were ap-plied to optimize the visualization of language areas.Resting-state fMRI data were processed by using an

in-house software, IClinfMRI [9], under MATLAB 2014a(The MathWorks, Inc., Natick, Massachusetts). The soft-ware calls functions in free for non-commercial use soft-ware including dcm2nii (https://www.nitrc.org/projects/dcm2nii/), AFNI (version 16.2.09) [10] and SPM12(v6685) (Welcome Department of Cognitive Neurology,Institute of Neurology, London, UK). The resting-statedatasets were pre-processed through slice timing, mo-tion correction, de-spiking, detrending, regressing outcovariates (including six motion parameters and two av-eraged fluctuations over masks of white matter and cere-brospinal fluid), band-pass filtering of 0.01–0.08 Hz, and4-mm FWHM smoothing. After the pre-processing,seed-based analysis was applied to detect the languagenetwork.

Kumar et al. Cancer Imaging (2020) 20:47 Page 2 of 9

https://www.nitrc.org/projects/dcm2nii/)https://www.nitrc.org/projects/dcm2nii/)

Two approaches were used for the seed selection: (1)by referencing to tb-fMRI activations (when available) or(2) by referencing a regional homogeneity (ReHo) map[11]. When using the first approach, peak tb-fMRI acti-vations in a primary language area apart from tumorwere used as seed locations. For example, when thetumor was close to the anterior language area, the seedwas selected in the posterior language area, and viceversa. When using the second approach, peaks of theReHo map within an anatomical mask of the anterior orposterior language network apart from tumor were usedas seed locations. The anatomical mask was generatedbased on the meta-analysis result obtained from Neuro-synth (http://neurosynth.org/) after the term “language”was entered [12]. The mask was constrained within theposterior inferior frontal gyrus, Broca’s area (pars trian-gularis and par opercularis), and middle frontal gyrus forthe anterior language area (ALA) and within angulargyrus, supramarginal gyrus, and superior and middletemporal gyri for the posterior language area (PLA) [13].The LONI Probabilistic Brain Atlas was used to applythe anatomical constraints [14]. When the ReHo mapwas used to help seed localization, we would attemptseeding at multiple local maxima within the anatomicalareas outlined by the meta-analysis mask. The final FCmap was qualitatively determined by (1) the existence ofpositive functional connectivity (FC) to the ALA or PLAof clinical interest and (2) minimizing non-specific FCsuch as in the CSF spaces. In addition, the presence ofFC in secondary language areas (i.e. language supple-mentary motor area and the visual word form area) in-creased diagnostic confidence in the final FC map. Thismanual searching process took on average 10 attemptsfor each case from which the final rs-fMRI FC map wasselected.Details in the implementation of the seed selection

methods are described in Hsu et al. [9]. As part of ourclinical SBC rs-fMRI post-processing workflow at thetime of this study, the tb-fMRI activation approach wasattempted first. If tb-fMRI activation was not availableor if this method of seeding was unsuccessful, then theregional homogeneity method was utilized. For eachseed location, a sphere of 6-mm radius was defined asthe seed region and a reference time course was gener-ated by averaging the time courses over the voxelswithin the region. The rs-fMRI connectivity map wascomputed using Pearson correlation between the refer-ence time course and that of each voxel in the brain(voxel size = 3.75 × 3.75 × 4mm3). The correlation coeffi-cient map was then converted to a Fisher’s z map. A Z-value threshold ranged from 0.6 to 1.0 was applied tooptimize the visualization of the language network. Foreach patient, multiple seeds were selected and each gen-erated a functional connectivity map. The final results

were determined by one of two clinical imaging physi-cists (P.H. and H-L.L.; both with more than 10 years ofclinical fMRI experience) who processed the rs-fMRIdata. Of note, > 1 mm translation or > 1 degree rotationin any direction for all post processed tb-fMRI and rs-fMRI cases were regarded as motion degraded.

Neuroradiology assessmentTwo neuroradiologists (V.A.K. and M.C.), with expertisein clinical fMRI, retrospectively reviewed the fMRIs frompatients whose tb-fMRI data were deemed limited and rs-fMRI data were subsequently post-processed. In these lim-ited tb-fMRI cases (1) the patient was too impaired to per-form tb-fMRI, (2) the patient performed poorly during tb-fMRI per clinician observation during scanning, (3) hadno to weak tb-fMRI BOLD signal specific to the languagearea of clinical interest near the tumor, (4) had nonspecificBOLD activation or (5) there was significant patient mo-tion during tb-fMRI acquisition as determined by motiongraphs. Nonspecific activation is defined as BOLD activa-tion outside the previously defined anterior or posteriorlanguage areas.To evaluate the benefit of SBC rs-fMRI for preoperative

functional language mapping, the neuroradiologists firstindependently reviewed the post-processed rs-fMRI data.A “yes” was recorded if the generated FC was within thepreviously defined ALA or PLA of clinical interest neartumor. A “no” was recorded, if the FC was outside the de-fined ALA or PLA of clinical interest, if any rs-fMRI arti-facts were observed, or for any other reason theneuroradiologist was not confident in the rs-fMRI results.If there was a discrepancy in agreement between the 2neuroradiologists; these cases were then reviewed togetherand a consensus “yes” or “no” was recorded.

Neurosurgery assessmentThe six referring neurosurgeons were provided a ques-tionnaire to assess the clinical value of rs-fMRI in thepreoperative mapping of language centers when tb-fMRIproved insufficient. The questionnaire included the fol-lowing question: When tb-fMRI was limited, based onyour knowledge of neuroanatomical functional areas, didyou find the rs-fMRI data “useful” in locating a potentialeloquent language area near tumor to assist direct cor-tical stimulation in this case? Neurosurgeon rating foreach case: 1) not useful; 2) neutral; 3) somewhat useful;4) definitely useful; or 5) not applicable.” The question-naire responses were collected and recorded.

Statistical analysisThe agreement between the two neuroradiologists’ rat-ing of the benefit of the post-processed rs-fMRI data forpresurgical language mapping was assessed usingCohen’s kappa statistic.

Kumar et al. Cancer Imaging (2020) 20:47 Page 3 of 9

http://neurosynth.org/

ResultsPatient demographics for whom resting-state fMRI waspost-processed are summarized in Table 1.

Causes of limited tb-fMRIOf the 134 patients with brain tumors who were evalu-ated for preoperative fMRI language mapping, 85 hadsuccessful tb-fMRI results and 49 had limited tb-fMRIresults. Thus, for these 49 patients, rs-fMRI was post-

processed. The following lists the causes of no or limitedtb-fMRI data: 1) poor patient performance as deter-mined by the neuropsychologist during the fMRI study(n = 12, 24.5%); 2) neuropsychological testing before thefMRI revealed that the patients were too impaired (e.g.,aphasic) to attempt tb-fMRI (n = 7, 14.3%); 3) weakBOLD activation in language area near tumor (N = 9;18.4%); 4) no BOLD activation near tumor (N = 8,16.3%); 5) nonspecific BOLD activation outside the pre-viously defined ALA or PLA (n = 11; 22.4%); 6) patientmotion artifact (n = 2; 4.1%). There were no tb-fMRItechnical (hardware or software) failures and no tb-fMRIfailures due to susceptibility artifacts.

Resting-state fMRI seed placementFor the 49 cases, the rs-fMRI seed location and resultantfunctional connectivity to the language area of clinicalinterest is summarized in Table 2.

Neuroradiological assessmentTwo neuroradiologists independently reviewed rs-fMRIdata from the 49 patients in which tb-fMRI was limited.One neuroradiologist determined that the post-processed rs-fMRI data was beneficial for language map-ping in 41 patients and the other determined that it wasbeneficial in 43 patients. Cohen’s kappa is 0.83, with a95% confidence interval (0.61, 1.00). With regard to the2 patients in whom there was a discrepancy in agree-ment, the first neuroradiologist felt that tb-fMRI sug-gested language reorganized to the contralateral ALA intwo patients. The other neuroradiologist felt that the rs-fMRI results were still of value in these two patients inthe event that reorganization was incomplete and re-sidual language function remained in the ALA area ipsi-lateral to the tumor. After consensus review, both

Table 1 Patient demographics for whom resting-statefunctional MRI was performed (n = 49)

Characteristic No. of patients

Mean age (range) 47.5 years (17–78 years)

Sex

Male 28

Female 21

Hand dominance

Right 45

Mixed 2

Left 2

Tumor location

Right hemisphere 1

Frontotemporal 1

Left hemisphere 48

Frontal 17

Parietal 7

Temporal 18

Insular 1

Frontoparietal 2

Temporoparietal 2

Intraventricular 1

Tumor enhancement

Enhancing 39

Non-enhancing 10

Surgery

Awake 37

Asleep 11

Canceled 1

Pathology

Grade II astrocytoma (WT:5; MUT:4; NOS:1) 10

Anaplastic astrocytoma (WT:2; MUT: 3) 5

Glioblastoma (WT:13; MUT: 8, NOS:1) 22

Oligodendroglioma 6

Metastasis 3

Pleomorphic xanthoastrocytoma 1

Anaplastic pleomorphic xanthoastrocytoma 2

WT Wild type, MUT Mutant, NOS Not otherwise specified

Table 2 Resting-state functional MRI findings (n = 49 patients)

No. of patients

Method of resting-state seed placement

Task-based BOLD activation 33

Regional homogeneity maps 16

Resting-state seed location ➔ functional connectivity location

Left ALA ➔ Left PLA 20

Left PLA ➔ Left ALA 19

Right PLA ➔ Right ALA 1

Right ALA ➔ Left ALA 2

Right ALA ➔ Left PLA 1

Resting-state functional MRI failure

Patient head motion 2

Nonspecific location of functional connectivity 1

Unknown system instability 3

ALA anterior langauge area, PLA posterior language area

Kumar et al. Cancer Imaging (2020) 20:47 Page 4 of 9

neuroradiologists deemed these 2 rs-fMRI cases helpfuland therefore a total of 43 cases were found beneficialfor language mapping.The neuroradiologists independently deemed that

the rs-fMRI data was not diagnostic and did notbenefit preoperative language mapping in 6 cases(12%). In two cases, the unsuccessful rs-fMRI was at-tributed to patient head motion during the rs-fMRIsequence acquisition. This was determined by review-ing the rs-fMRI motion graphs. In one patient, theFC obtained was in a nonspecific anatomic locationoutside the previously defined PLA. The remaining 3cases of rs-fMRI failure were due to an unknown sys-tem instability.

Neurosurgical assessmentThe six cases deemed rs-fMRI failures by both neurora-diologists were not included for neurosurgery assess-ment. The neuroradiologists submitted the remaining 43rs-fMRI FC results for neurosurgery assessment. The sixneurosurgeons each reviewed the rs-fMRI FC results fortheir own patients. The neurosurgeons found the post-processed rs-fMRI results to be “definitely” useful in 26patients (60%) and “somewhat” useful in 13 patients(30%) for localizing potential anterior and posterior lan-guage areas and thus providing potential stimulationpoints near tumor to assist direct cortical stimulation.For two patients (5%), a neutral rating was rendered. A“not applicable” rating was given for two patients (5%).In the first, the surgery was canceled. In the other, rs-fMRI showed good ALA FC, but this was not of interestto the surgeon because he had no plan to resect nearthis area. There were no negative ratings of the rs-fMRIdata by the neurosurgeons.

Illustrative casesCase 1: successful rs-fMRI after no tb-fMRI activationIn a 24-year-old right-handed woman with an IDHmutant recurrent left frontal glioblastoma, tb-fMRI wasperformed but showed no BOLD activation in the leftanterior language area; therefore, rs-fMRI was post-processed. The tb-fMRI activation in left Wernicke’sarea was used as a seed for rs-fMRI. Functional connect-ivity was successfully found in left Broca’s area (Fig. 1).

Case 2: successful rs-fMRI after limited tb-fMRI due topatient motionIn a 58-year-old right-handed woman diagnosed with anIDH wild-type left temporal glioblastoma, tb-fMRI wasperformed but was limited due to extensive patient mo-tion; therefore, rs-fMRI was post-processed. A seedguided by ReHo was placed in left Broca’s area to obtainFC in left Wernicke’s area which was posterior to theglioma (Fig. 2).

Case 3: successful rs-fMRI after poor patient performanceduring tb-fMRIIn a 73-year-old left-handed man with an IDH wild-typeglioblastoma in the left posterior superior temporalgyrus, the patient could not follow instructions duringtb-fMRI and therefore rs-fMRI was post-processed. Re-gional homogeneity was used to guide seed placement inleft Broca’s area which demonstrated FC to the left infer-ior parietal lobule (Geschwind’s area) above the glioma(Fig. 3).

Case 4: successful rs-fMRI with seed placement in thecontralateral hemisphereIn a 52-year-old right-handed man with an IDH mutantleft insular grade II astrocytoma, tb-fMRI could not be

Fig. 1 Successful resting-state fMRI after no task-based fMRI activation. Post-contrast axial T1 image (a) shows a left frontal glioblastoma (arrow).Task-based fMRI (b) demonstrates no BOLD activation in left anterior language area and robust BOLD activation related to the sentencecompletion paradigm (yellow) and category fluency paradigm (pink) in left Wernicke’s area (curved arrow). Resting-state fMRI (c) with seedplacement in Wernicke’s area at site of task-based BOLD activation (dashed arrow) results in robust functional connectivity to left Broca’sarea (arrowhead)

Kumar et al. Cancer Imaging (2020) 20:47 Page 5 of 9

performed due to patient impairment by tumor; there-fore, rs-fMRI was post-processed. Initial seed placementin the left anterior language area did not elicit FC to theleft posterior language area. Regional homogeneity wasused to guide seed placement in the contralateral rightposterior inferior frontal lobe, which demonstrated sym-metric FC in the bilateral Wernicke’s areas. In the leftWernicke’s area, FC was noted abutting the FLAIR ab-normality (Fig. 4).

Case 5: unsuccessful rs-fMRI due to patient motionIn a 62-year-old right-handed man with colorectal can-cer who had a left frontal brain metastasis, tb-fMRI wasattempted but was limited due to weak BOLD activationin the ALA; therefore, rs-fMRI was post-processed. Aseed was placed at the tb-fMRI BOLD activation in theleft PLA. rs-fMRI demonstrated potential FC posterolat-eral to the glioma. However, review of the sagittal and

coronal rs-fMRI showed that FC was visualized only in asingle plane, and therefore the rs-fMRI was deemednon-diagnostic. The single slice FC artifact was due topatient motion during the rs-fMRI acquisition. Patientmotion can affect global FC (Fig. 5).

DiscussionWe evaluated the usefulness of resting-state fMRI forpresurgical language mapping when task-based fMRIwas limited in 49 patients with brain tumors. A seed-based correlation (SBC) approach used either tb-fMRIlanguage activations apart from tumor or a regionalhomogeneity (ReHo) map to obtain functional connect-ivity (FC) to language areas of interest. Two neuroradiol-ogists independently found the post-processed rs-fMRIdata using SBC analysis was beneficial for functional lan-guage mapping in 41 (84%) and 43 (88%) cases respect-ively; Cohen’s kappa is 0.83, with a 95% confidence

Fig. 2 Successful resting-state fMRI after patient motion limited task-based fMRI. Post-contrast axial T1 image (a) shows a left temporalglioblastoma (arrow). Task-based fMRI (b) displays nonspecific BOLD activations (i.e. yellow = sentence completion; blue = letter fluency) and noisedue to patient motion. This case illustrates a regional homogeneity functional connectivity map confined to the Broca’s area meta-analysis (c)showing 4 seed candidates. Seed #1 results in the final resting-state fMRI functional connectivity map (d) which demonstrates robust connectivityin left Wernicke’s area (dashed arrow)

Fig. 3 Successful resting-state fMRI after poor patient performance during task-based fMRI. Post-contrast sagittal T1 image (a) shows a leftposterior temporal glioblastoma (arrow). Axial resting-state fMRI (b) with seed placement in left Broca’s area employing the regional homogeneitymethod (arrowhead). Sagittal resting-state fMRI (c) shows functional connectivity in the left inferior parietal lobule (dashed arrow) superior tothe tumor

Kumar et al. Cancer Imaging (2020) 20:47 Page 6 of 9

interval (0.61, 1.00) when tb-fMRI was limited. A subse-quent consensus review by the two neuroradiologistsfound a total of 43 cases beneficial for language map-ping. We found a 12% rs-fMRI failure rate for languagemapping which was somewhat similar to a study using amultilayer perceptron rs-fMRI analysis which showed ars-fMRI failure rate of 13% for combined motor and lan-guage mapping [15].Further, six neurosurgeons found that the rs-fMRI re-

sults were “definitely useful” (26 cases; 60%) and “some-what useful” (13 cases: 30%) in locating the ALA or PLAof clinical interest near tumor in their own cases. For ex-ample, one neurosurgeon commented in the question-naires that the rs-fMRI results are “useful” in patientsthat are too impaired by tumor to perform substantiveintraoperative direct cortical stimulation and have no tb-fMRI results. The rs-fMRI results serves as the onlymethod of language brain mapping and could highlighta potential location to avoid (if possible) or approachwith more caution during tumor resection.

In patients who performed poorly on speech para-digms during tb-fMRI (n = 12) and who were too im-paired to even attempt tb-fMRI (n = 7); the postprocessed rs-fMRI FC was deemed beneficial by bothneuroradiologists in 89% of cases for language mapping.It is in this subset of patients, that implementation ofclinical rs-fMRI would be especially valuable. In in-stances when rs-fMRI showed robust FC in the same lo-cation as weak tb-fMRI activation (6 of 9 cases), thisconcordance increased diagnostic confidence in thelocalization of the language center. Therefore, rs-fMRIcan also be used to verify equivocal tb-fMRI results.When FC was observed in only one slice (Case/Fig. 5);

the rs-fMRI data was considered non-diagnostic in all 6(12%) cases. This unique rs-fMRI artifact is due to anundesired instability (i.e. patient motion) which canaffect global FC. We acquire the rs-fMRI as the firstfMRI sequence prior to all task fMRI paradigms to miti-gate the chances of patient head motion. In addition, itis important to acquire the rs-fMRI sequence first toavoid the potential influence of cognitive tasks on theresting state networks.Methods used to identify resting-state networks are

classified into data-driven and hypothesis-drivenmethods. Resting state-ICA is a popular data-drivenmethod which can display multiple networks simultan-eously; however, it may not display the network of clin-ical interest. Further in practice, the network identifiedby ICA, if present, must be individually selected out bythe investigator [6]. Resting-state SBC analysis is ahypothesis-driven method and more intuitive in target-ing a specific language functional area of interest; how-ever, a priori target must be determined, and thisvaries between studies [16]. In 19 patients, Cochereauet al. showed a significant correlation between lan-guage networks identified with SBC rs-fMRI and elo-quent language cortex confirmed with direct corticalstimulation [17].In 67% of the patients for whom SBC rs-fMRI was

used in our study, tb-fMRI activation in the primary

Fig. 4 Successful resting-state fMRI with seed placement in thecontralateral hemisphere. Resting-state fMRI (a) with seed in thecontralateral right Broca’s area (arrowhead) using the regionalhomogeneity method in a patient with an infiltrating left insularglioma. Resting-state fMRI (b) shows functional connectivity in thebilateral Wernicke’s areas (dashed arrows)

Fig. 5 Unsuccessful resting-state fMRI due to patient motion. Axial resting-state fMRI (a) demonstrates possible functional connectivity (dashedarrow) posterior lateral to a left frontal brain metastasis. Coronal (b) and sagittal (c) resting-state fMRI shows functional connectivity only in asingle plane (arrowheads) due to patient motion. This was deemed a resting-state fMRI failure

Kumar et al. Cancer Imaging (2020) 20:47 Page 7 of 9

language area apart from tumor was used to guide seedplacement. In the other 33% of patients, we used theReHo method [18] to assist in seed placement for SBCrs-fMRI analysis. We did not have any cases in whichthe tumor or tumor mass effect distorted the anatomy tothe degree that it affected rs-fMRI seed placement usingtb-fMRI activations or ReHo. Anatomic landmarks alonewere not used for seed placement in any of the casesdue to known issues related to anatomic distortioncaused by tumor or tumor mass effect. In a recentlypublished research study performed at our institution,we compared SBC rs-fMRI maps generated using tb-fMRI activations (apart from tumor) as seed points, anautomated ReHo method, and a canonical (anatomicallandmark) approach. The ReHo method yielded rs-fMRIlanguage mapping results that were in greater agreementwith the results of tb-fMRI, with significant higher Dicecoefficients (p < 0.5) than that of tb-fMRI and canonicalapproaches within the putative language areas [19].Based on the results of this study, we plan to formallyintegrate an automated ReHo method to guide SBC rs-fMRI seed placement into our clinical workflow in thenear future.In 40 patients, FC was obtained in the hemisphere ip-

silateral to tumor. Interestingly, in 3 patients in whichFC could not be obtained from ipsilateral seed place-ment, the ReHo method was used to guide seed place-ment in the contralateral ALA to obtain successful FCto the ipsilateral to tumor ALA (2 cases) and ipsilateralto tumor PLA (1 case). Therefore, we have also shownin this study that contralateral ALA or PLA seeding is apotential viable secondary method of obtaining FC tothe ipsilateral primary language areas of interest neartumor.

LimitationsThe number of limited tb-fMRI may be consideredhigher than expected due to inclusion of 6 weak tb-fMRIcases which were found to directly correlate with rs-fMRI FC results. Comparing tb-fMRI and rs-fMRI fail-ure rate was not the aim of this study. The rs-fMRI wasthe first sequence obtained and we employed a closedeye rather than open eye technique to reduce potentialeye strain for the subsequent tb-fMRI paradigms. We donot feel the patient falling asleep during the rs-fMRI ac-quisition is a confounding factor due to the use of aclosed eye technique. In fact, there was no patient thathad fallen asleep prior to the start of the tb-fMRIparadigms.We did not investigate the potential reorganization of

language function to the contralateral hemisphere thatmay occur in rare cases. Our goal was to identify a po-tential language area near the tumor using rs-fMRI. Fur-ther studies are needed to determine if rs-fMRI can

reliably detect contralateral hemisphere languagereorganization. We found seeding either the ALA orPLA resulted in co-activation of both ipsilateral andcontralateral primary language networks in most cases.This coincides with findings from Doucet et al. [20],who found that SBC rs-fMRI identified broader andmore bilateral language regions than did tb-fMRI.Therefore, at this time, we cannot determine hemi-spheric dominance with our current SBC rs-fMRI tech-nique. We found a study by Lou et al. [21] whichdemonstrated that seed placement over the bilateral pre-supplementary motor area (pre-SMA) allowed identifica-tion of left language lateralization in 30 right-handedhealthy patients. We did not use the bilateral pre-SMAas a seed and may utilize this area for future seed place-ment. Even without knowledge of hemispheric domin-ance, SBC rs-fMRI in this study could locate potentialALA and PLA functional language areas ipsilateral totumor.While the neurosurgical assessment compliments the

neuroradiology assessment regarding the value of theSBC rs-fMRI technique used in this retrospective study;direct cortical stimulation data, percentage of tumor re-section, complication rates including postoperative lan-guage deficits, and other clinical outcome data would bemore informative. Lemée et al. found ICA rs-fMRI wasable to detect language areas which correlated with cor-tical mapping with a sensitivity of 100%, compared to65.6% with tb-fMRI [22]. We are currently undertakinga similar prospective analysis comparing SBC rs-fMRI,tb-fMRI, and cortical mapping at our institution.

ConclusionsThis study serves as a proof of concept that SBC rs-fMRI can be valuable for clinical preoperative languagemapping when a patient cannot perform tb-fMRI or ifthe results are limited.

AbbreviationsALA: Anterior language area; BOLD: Blood oxygenation level-dependent;FC: Functional connectivity; ICA: Independent component analysis;PLA: Posterior language area; ReHo: Regional homogeneity; rs-fMRI: Resting-state fMRI; SBC: Seed-based correlation; tb-fMRI: Task-based fMRI

AcknowledgementsNot applicable.

Authors’ contributionsVAK, H-LL devised study concept. VAK, IMH wrote the manuscript. SP, MMC,PH, KN, SDF, GR, FFL provided data for study and edited manuscript; ALY, RC,JJ, DFS; edited manuscript. All authors read and approved the finalmanuscript.

FundingNot applicable.

Availability of data and materialsNo additional data.

Kumar et al. Cancer Imaging (2020) 20:47 Page 8 of 9

Ethics approval and consent to participateAll procedures performed in studies involving human participants were inaccordance with the ethical standards of the Helsinki declaration. Ourinstitutional review board approved this retrospective study (PA19–0417)with waiver of informed consent.

Consent for publicationNot applicable.

Competing interestsThe authors declare that they have no completing interests.

Author details1Department of Neuroradiology, The University of Texas MD AndersonCancer Center, Houston, TX, USA. 2Department of Neurosurgery, TheUniversity of Texas MD Anderson Cancer Center, Houston, TX, USA.3Department of Diagnostic Radiology, The University of Pittsburgh MedicalCenter, Pittsburgh, PA, USA. 4Department of Imaging Physics, The Universityof Texas MD Anderson Cancer Center, Houston, TX, USA. 5Department ofNeuro-Oncology, The University of Texas MD Anderson Cancer Center,Houston, TX, USA.

Received: 19 March 2020 Accepted: 2 July 2020

References1. Petrella JR, Shah LM, Harris KM, et al. Preoperative functional MR imaging

localization of language and motor areas: effect on therapeutic decisionmaking in patients with potentially resectable brain tumors. Radiology.2006;240(3):793–802.

2. McGirt MJ, Chaichana KL, Gathinji M, et al. Independent association ofextent of resection with survival in patients with malignant brainastrocytoma. J Neurosurg. 2009;110(1):156–62.

3. Lee MH, Smyser CD, Shimony JS. Resting-state fMRI: a review of methodsand clinical applications. AJNR Am J Neuroradiol. 2013;34(10):1866–72.

4. Shen HH. Core Concept: resting-state connectivity. Proc Natl Acad Sci U S A.2015;112(46):14115–6.

5. Smith SM, Fox PT, Miller KL, et al. Correspondence of the brain’s functionalarchitecture during activation and rest. Proc Natl Acad Sci U S A. 2009;106(31):13040–5.

6. Tie Y, Rigolo L, Norton IH, et al. Defining language networks from resting-state fMRI for surgical planning—a feasibility study. Hum Brain Mapp. 2014;35:1018–30.

7. Branco P, Seixas D, Deprez S, et al. Resting-state functional magneticresonance imaging for language preoperative planning. Front HumNeurosci. 2016;10:11.

8. Rosazza C, Minati L, Ghielmetti F, Mandelli ML, Bruzzone MG. Functionalconnectivity during resting-state functional MR imaging: study of thecorrespondence between independent component analysis and region-of-interest-based methods. AJNR Am J Neuroradiol. 2012;33(1):180–7.

9. Hsu A-L, Hou P, Johnson JM, et al. IClinfMRI software for integratingfunctional MRI techniques in presurgical mapping and clinical studies. FrontNeuroinform. 2018;12:11.

10. Cox RW. AFNI: software for analysis and visualization of functional magneticresonance neuroimages. Comput Biomed Res. 1996;29(3):162–73.

11. Zang Y, Jiang T, Lu Y, He Y, Tian L. Regional homogeneity approach to fMRIdata analysis. Neuroimage. 2004;22(1):394–400.

12. Yarkoni T, Poldrack RA, Nichols TE, et al. Large-scale automated synthesis ofhuman functional neuroimaging data. Nat Methods. 2011;8(8):665–70.

13. Binder JR. Current controversies on Wernicke’s area and its role in language.Curr Neurol Neurosci Rep. 2017;17(8):58.

14. Shattuck DW, Mirza M, Adisetiyo V, et al. Construction of a 3D probabilisticatlas of human cortical structures. Neuroimage. 2008;39(3):1064–80.

15. Leuthardt EC, Gloria G, Kathleen Bandt S, et al. Integration of resting-statefunctional MRI into clinical practice—a large single institution experience.PLoS One. 2018;13(6):e0198349.

16. Wongsripuemtet J, Tyan AE, Carass A, et al. Preoperative mapping of thesupplementary motor area in patients with brain tumor using resting-statefMRI with seed-based analysis. AJNR Am J Neuroradiol. 2018;39(8):1493–8.

17. Cochereau J, Deverdun J, Herbet G, et al. Comparison between resting statefMRI networks and responsive cortical stimulations in glioma patients. HumBrain Mapp. 2016;37(11):3721–32.

18. Yan FX, Wu CW, Cheng SY, et al. Resting-state functional magneticresonance imaging analysis with seed definition constrained by regionalhomogeneity. Brain Connect. 2013;3(4):438–49.

19. Hsu A-L, Chen HS-M, Hou P, et al. Presurgical resting-state functional MRIlanguage mapping with seed selection guided by regional homogeneity.Magn Reson Med. 2019. https://doi.org/10.1002/mrm.28107.

20. Doucet GE, He X, Sperling MR, et al. From “rest” to language task: taskactivation selects and prunes from broader resting-state network. HumBrain Mapp. 2019;38(5):2540–52.

21. Lou W, Peck KK, Brennan N, Mallela A, Holodny A. Left-lateralization ofresting state functional connectivity between the presupplementary motorarea and primary language areas. Neuroreport. 2017;28(10):545–50.

22. Lemée JM, Berro DH, Bernard F, et al. Resting-state functional magneticresonance imaging versus task-based activity for language mapping andcorrelation with perioperative cortical mapping. Brain Behav. 2019;9(10):e01362.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Kumar et al. Cancer Imaging (2020) 20:47 Page 9 of 9

https://doi.org/10.1002/mrm.28107

AbstractBackgroundMethodsResultsConclusions

BackgroundMethodsSubjectsImaging techniquesMRI acquisitionTb-fMRI paradigmsWorkflow and processing pipeline (tb-fMRI & rs-fMRI)

Neuroradiology assessmentNeurosurgery assessmentStatistical analysis

ResultsCauses of limited tb-fMRIResting-state fMRI seed placementNeuroradiological assessmentNeurosurgical assessmentIllustrative casesCase 1: successful rs-fMRI after no tb-fMRI activationCase 2: successful rs-fMRI after limited tb-fMRI due to patient motionCase 3: successful rs-fMRI after poor patient performance during tb-fMRICase 4: successful rs-fMRI with seed placement in the contralateral hemisphereCase 5: unsuccessful rs-fMRI due to patient motion

DiscussionLimitations

ConclusionsAbbreviationsAcknowledgementsAuthors’ contributionsFundingAvailability of data and materialsEthics approval and consent to participateConsent for publicationCompeting interestsAuthor detailsReferencesPublisher’s Note